Rapid Model Development

This research develops innovative techniques for updating 3D representations of targets used for SAR ATR, e.g., of the type being employed on DARPA's MSTAR program, directly from data. The work will provide a critical bridge between models and data, allowing image exemplars of modified targets to be of immediate use in updating target representations. The proposed work would also extend naturally to address denied target problems for which the models must be identified entirely from remote sensing data. We focus our Phase I program on assessing the feasibility of representing targets with collections of updateable parameterized statistical scattering primitives. The approach will be evaluated both theoretically and with an empirical study which will attempt to model representation to better reflect real data exemplars. The Phase I evaluation will also benchmark the time and cost of our model construction, which we expect to be quite favorable. The Phase II effort would then extend Phase I by enriching the set of descriptors in the core representation, producing five improved target models, inserting them into MSTAR, and addressing the denied target problem for which no prior model is available.